Cloud Migration Strategy: Plan, Phases, Risks, Checklist

A cloud migration strategy is the decision, made workload by workload, about what moves to the cloud, how it moves, in what order, and what happens to it afterward. The strategies that hold up share three habits: they measure before they move, they pick a different approach for different applications, and they migrate in waves with a rollback plan written down before the first cutover. This guide walks through each of those, from the migration models to the checklist, so you can write a plan your IT team, finance team and leadership can all sign off on.
If your move involves custom software, legacy systems or AI workloads, the strategy has a second job: deciding which applications deserve to be rebuilt and which only need to be relocated. We cover that distinction below.
TL;DR
- A strategy is a per-workload decision, not a single method for the whole estate. The common options are the 7 Rs: retire, retain, rehost, relocate, repurchase, replatform and refactor.
- Inventory and dependency mapping come first. Anything you have not measured, you cannot plan, price or sequence.
- Move in waves: foundation first, low-risk workloads next, business-critical systems after early wins. Write rollback criteria before any cutover.
- Migration is not finished at cutover. Validate costs, backups, restores and security against your pre-migration baseline, then review on a schedule.
- Budget for more than infrastructure fees. Labor, testing and modernization often decide the real cost.
Table of Contents
- What a cloud migration strategy actually contains
- The 7 Rs: choosing a migration approach for each workload
- Step one: assess what you have
- Step two: build the plan
- The phases of cloud migration
- Cloud migration risks and how to reduce them
- Cloud migration checklist
- A worked example: a mid-sized company with 40 applications
- Where we can help
- Our perspective
- FAQ
- Sources
What a cloud migration strategy actually contains
A cloud migration strategy is a business case, a scope, and a method for moving each existing IT component to a new cloud environment. AWS describes it as outlining every existing component and the best way to migrate each one. Notice what that definition leaves out: it does not say “move everything.” Some components are better off retired or left where they are.
In practice, a usable strategy document answers five questions:
- Why are we moving? Cost, resilience, speed of delivery, end of hardware life, a data center exit, or a need to run AI workloads that on-premises systems cannot support. Pick the one or two that matter most, because they drive every later trade-off.
- What are we moving? A catalog of applications, data stores, and the infrastructure under them.
- How does each item move? One of the Rs below.
- In what order and by whom? Waves, owners, timelines.
- How will we know it worked? Measurable targets for cost, performance, availability and recovery, set against a baseline captured before the move.
The last question is the one most plans skip, and it is the reason many finished migrations cannot prove they paid off.
The 7 Rs: choosing a migration approach for each workload
You will see this framework called the 5 Rs, 6 Rs or 7 Rs. They are the same idea at different stages of growth. The model began with Gartner’s five Rs in 2010 (rehost, refactor, revise, rebuild, replace). Retire was added to make six, and retain made seven, recognizing that not every application or dataset can or should move, for reasons such as legal or privacy restrictions, cost or latency.
The seven, using the AWS definitions:
| Strategy | What it means | Typical fit |
|---|---|---|
| Retire | Switch off an application nobody needs | Duplicates, unused tools found during inventory |
| Retain | Leave it where it is for now | Regulated data, recent hardware investment, latency-sensitive systems |
| Rehost (“lift and shift”) | Move servers and workloads with minimal changes | Fast exits from a data center, stable apps with no urgent need to change |
| Relocate | Move virtualized workloads between platforms without rebuilding | Hypervisor-based estates |
| Repurchase (“drop and shop”) | Replace the application with a cloud-native or SaaS product | CRM, HR, email, commodity tools |
| Replatform (“lift, tinker and shift”) | Small optimizations without changing core architecture | Moving a self-managed database to a managed service |
| Refactor (re-architect) | Redesign the application for cloud-native capabilities | Core products that must scale, or integrate AI and real-time data |
Which R to start with
The AWS Prescriptive Guidance for large migrations recommends rehost, replatform, relocate and retire as the common strategies in big programs. It advises against refactoring during a large migration because it is the most complex option, and suggests modernizing applications after migration is complete.
That guidance translates into a simple rule: move first, modernize second, except where the application itself is the reason for the project. If a system is brittle, undocumented or cannot support the features your business now needs, lifting it unchanged simply relocates the problem. For those few systems, a scoped modernization is cheaper than paying cloud rates for something you will rebuild in a year. Our guide on per-app legacy modernization shows how to test that decision on one application before committing to many.
Step one: assess what you have
The AWS framework puts it bluntly: you cannot effectively move what you do not measure. Assessment is where migrations are won or lost, and it is more concrete than most plans suggest.
The Microsoft Cloud Adoption Framework tells teams to collect, for each workload:
- Baseline performance metrics: CPU, memory, disk I/O, network throughput, response times and service-level metrics.
- Configuration and licensing details: VM sizes, operating systems, storage and software licenses, since licensing can change the cost picture.
- Security and identity configuration: who and what can access the workload today.
- Compatibility issues: found with discovery tools, then listed as pre-migration blockers or post-migration items.
- Dependencies: internal and external, confirmed with the people who own the workload.
Google Cloud’s migration planning documentation describes the same sequence: build a catalog of workloads, map them to infrastructure, map dependencies, and identify a high-level migration path for each. It also recommends running a migration risk assessment in parallel, so risks are measured and mitigations start early.
Two practical points from experience of how these assessments go wrong:
- Dependencies hide in the unofficial parts. A reporting job that reads from a database nightly, a partner integration on a fixed IP address, a licensing server in a closet. Ask workload owners, not just scanners.
- Not everything will go. Some systems have to stay near your premises for latency, sovereignty or equipment reasons. If part of your estate stays, plan the on-site layer deliberately; building a hybrid cloud infrastructure quickly is a legitimate part of the strategy, not a failure of it.
If data location rules apply to you, settle them at this stage. Our walkthrough of data residency requirements turns a legal obligation into technical evidence you can show an auditor, which determines which regions and services you are allowed to use.
Check readiness beyond the technology
Microsoft’s guidance also calls for a readiness assessment of your team’s skills, and notes that outside expertise can validate the strategy, recommend tools and set realistic timelines. AWS evaluates readiness across six perspectives: business, people, governance, platform, security and operations. A migration with excellent technical planning and no named owners still stalls.
Step two: build the plan
A strategy says what to do with each workload. A plan says when. According to the Cloud Adoption Framework, a migration plan defines the specific order, timing and approach for moving workloads, turning high-level strategy into a deployment sequence.
The same guidance gives sequencing rules that are easy to apply:
- Rank workloads by business value against effort. High value and low effort go first, because quick wins build confidence and funding.
- Start with simpler workloads and move non-production environments before production.
- Schedule critical systems after the early waves have succeeded.
- Include one or two representative complex workloads in each early wave so you meet the hard scenarios while the stakes are still low.
A wave structure you can adapt
| Wave | Contents | Purpose |
|---|---|---|
| 0 | Landing zone: accounts, networking, identity, security baseline, cost tagging | The foundation every later wave depends on |
| 1 | Low-risk and non-production workloads, plus a pilot | Prove the process and tooling |
| 2 | Supporting services and moderately complex apps | Build speed using lessons from wave 1 |
| 3 | Business-critical applications | Move with a rehearsed cutover and rollback |
| 4 | Decommissioning of old systems after parallel-run validation | Stop paying twice |
The exact number of waves depends on how many applications you have. A small company may need two or three. A data center exit with hundreds of applications needs more, and a longer timeline.
Write the rollback plan now
The Cloud Adoption Framework states that rollback criteria and procedures must be established before any migration or deployment begins. A rollback plan lets you reverse a failed change quickly and limit downtime. Test one thing: could someone other than the lead engineer run it at 2 a.m.? If the plan depends on the one person who built the system, it is not a plan.
The phases of cloud migration
Different frameworks name the phases differently, but the order is the same. Google Cloud describes a multi-phase process of assessment, planning, migration, and optimization and modernization. Microsoft’s migrate overview uses prepare, assess, deploy and release. Combined:
- Prepare and align. Agree the goals, assign owners, confirm skills and decide on governance. Nothing technical should start until someone owns each workload.
- Assess. Build the inventory, baselines and dependency map described above. Define security and compliance requirements and determine which workloads are eligible for the cloud.
- Plan. Choose the R for each application, design the target infrastructure, and sequence the waves.
- Migrate (deploy). Execute in staged waves, validate each workload in the cloud, and apply lessons learned to the next wave.
- Release and operate. Test, optimize, document, and hand workloads to operations.
- Optimize and modernize. Monitor usage and cost, adopt managed services, and refactor where it pays off.
Cloud migration risks and how to reduce them
Migration risk is workload-specific. Google Cloud’s best-practice guidance points to downtime tolerance, interdependencies, configuration changes and tested rollback strategies as the factors that should shape the plan and every execution step. Here are the risks that most often hurt, and the matching control for each.
| Risk | What goes wrong | Control |
|---|---|---|
| Missed dependencies | An unmapped integration breaks after cutover | Dependency mapping with workload owners; a pilot wave |
| Weak validation | Defects reach production users | Functional, performance and security testing before cutover |
| No tested rollback | A failed cutover becomes an outage | Written rollback criteria, rehearsed by more than one person |
| Security gaps during the first move | Something is left open by accident | Use the platform’s security dashboards and resolve warnings |
| Cost surprises | Bills exceed the business case | Baseline costs, tagging, budget alerts and a post-move review |
| Over-ambitious scope | Refactoring inside a large migration delays everything | Move first, modernize after, except for systems that justify it |
On the security row: the UK National Cyber Security Centre’s lift-and-shift guidance notes that mistakes are most likely during the first move to the cloud, that security dashboards can flag anything left insecure, and that warnings should be resolved unless you are certain they are harmless. It is a UK government source, but the principle applies anywhere.
Cost deserves its own attention because it is where finance teams feel risk first. Hosting fees are only part of the bill. Labor, testing, parallel running and training add up, and our analysis of cloud migration cost for IT and finance shows labor can reach 55% of the total, which is worth building into the business case from the start.
Cloud migration checklist
Use this as a working list. It follows Google Cloud’s workload migration checklist and the Microsoft execution guidance cited above.
Before the move
- Name stakeholders and an owner for each workload.
- Inventory applications, data stores, users and capacity.
- Document dependencies, service-level agreements, and security and compliance requirements.
- Record baseline performance and current run costs.
- Choose an R for each application and design the target environment.
- Test the migration tools and the target’s performance.
- Write migration steps, cutover timing and rollback criteria.
During the move
- Migrate in phases, one wave at a time.
- Validate each workload in the cloud before you rely on it.
- Apply lessons from each wave to the next.
- Monitor workload behavior and cloud usage.
- For the final cutover, Microsoft’s execution guidance lists a final data synchronization, integrity checks, core-function testing, traffic redirection, and thorough post-cutover validation and monitoring.
After the move
- Compare costs with the pre-migration baseline and investigate changes at the resource level.
- Confirm metrics, logs and alert rules are working.
- Verify backup jobs and test restores against your recovery time and recovery point objectives (how long you can be down, and how much data you can lose).
- Resolve security recommendations and collect user feedback.
- Schedule regular reviews of cost, performance, reliability, security and operations.
The Cloud Adoption Framework’s post-migration guidance covers each of these checks. Teams that skip them often discover over-provisioned servers and silent backup failures only when something breaks or the invoice arrives. Our guide to cloud cost optimization picks up where the migration ends.
A worked example: a mid-sized company with 40 applications
Take a hypothetical company with 40 applications, a single on-premises data center and a lease ending in 18 months. This is an illustration, not a client case.
The inventory shows six applications nobody has opened in a year: retire. Four are commodity tools (email archive, expense system, internal wiki, HR): repurchase as SaaS. Two ERP-linked systems depend on a local device and a regulator’s location rule: retain for now. Twenty-two stable internal applications: rehost or replatform in waves 1 and 2. Six customer-facing and core systems remain. Of those, four are rehosted and modernized later, and two are old, hard to change and blocking a planned AI feature, so they are refactored, each starting with a small proof of concept.

The result is a plan in which only a handful of applications absorb the expensive work, while the majority move quickly. That is what a per-workload strategy buys you.
Where we can help
If your migration includes custom software, legacy applications or AI, we can help you scope it. Our cloud infrastructure services cover AWS, Azure and GCP, and our approach is technology-agnostic. A proof-of-concept on one application, from €5,000 one-off, is a practical way to test a refactor decision before committing the whole estate.
For larger programs, we work through two engagement models listed on our cooperation page: an agile engagement at €5,000 to €40,000 per month, or a fixed-price turnkey project at €5,000 to €60,000 one-off.

Our perspective
Most migration plans we read treat the cloud as the destination. We treat it as the starting line. The point of moving is that your systems can then do something they could not do before: scale on demand, share data across teams, and support automation and AI decisions in real time.
That is why we integrate AI into core systems rather than bolting it on later, and why we think the choice of R matters so much. A rehost is the right call for a stable application. For the few systems that carry your growth, a lift-and-shift only moves the constraint into a more expensive place.
We stay technology-agnostic, so the target platform follows your workloads, not our preferences. We also insist on measurable results: a baseline before the move, targets in the plan, and a review after cutover that compares the two. We prefer long-term partnerships because the real savings and the real improvements show up in the months after the migration, not on cutover day.
Matija
FAQ
What are the 6 cloud migration strategies?
The six Rs are rehost, replatform, repurchase, refactor, retire and retain. The earlier five-R model from 2010 did not include retire, and the seven-R model adds relocate for virtualized workloads. In every version, you choose the approach per application.
How long does a cloud migration take?
It depends on how many applications you have and how tangled their dependencies are. A small estate may take a few months, while a full data center exit with many applications is a multi-quarter program. Pilot waves and early dependency mapping are what keep timelines predictable.
Should I refactor during the migration?
For most applications, move first and modernize afterward, as the AWS large-migration guidance recommends. Refactor during the move only when the application is blocking a business goal, such as scaling or adding AI features, and test the decision on a small proof of concept first.
What should a migration plan include at minimum?
A workload inventory with owners, dependency maps, a chosen strategy per application, a wave sequence, security and compliance requirements, rollback criteria, and measurable targets against a baseline. Add a post-migration review date so the work does not end at cutover.
Sources
- What is a Cloud Migration Strategy? (AWS)
- Understanding the 7 Rs of cloud migration strategies (AWS Builder Center)
- About the migration strategies (AWS Prescriptive Guidance)
- Assess your workloads for cloud migration (Microsoft Learn)
- Plan your migration (Microsoft Learn)
- Migrate overview (Microsoft Learn)
- Execute migration to cloud (Microsoft Learn)
- Optimize workloads after migration (Microsoft Learn)
- About migration planning (Google Cloud Migration Center)
- What is cloud migration? (Google Cloud)
- Migrate workloads to the public cloud checklist (Google Cloud)
- Migration to Google Cloud best practices
- How to lift and shift successfully (UK NCSC)
- Hybrid cloud infrastructure: building the on-site layer





